Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Get rid of TableRegressionModel by storing formula directly in model #339

Merged
merged 18 commits into from
Nov 20, 2022
Merged

Get rid of TableRegressionModel by storing formula directly in model #339

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
18 commits
Select commit Hold shift + click to select a range
5df0fb5
Get rid of TableRegressionModel by storing formula directly in model
nalimilan Oct 3, 2019
d3f9430
Cleanup
nalimilan Jun 19, 2022
b670a14
Reuse existing test instead of adding another one
nalimilan Jun 19, 2022
a83cb1b
Tables.jl fixes
nalimilan Jun 19, 2022
bd00e51
Fixes
nalimilan Jun 19, 2022
c244147
Docs, etc.
nalimilan Jun 19, 2022
fc0704a
Docs, contrasts
nalimilan Jun 19, 2022
399979f
Review fixes, more tests
nalimilan Sep 11, 2022
8365831
Merge remote-tracking branch 'origin/master' into nl/tableregressionm…
nalimilan Sep 11, 2022
1a012f1
Fix doctests on Julia 1.8
nalimilan Sep 11, 2022
f9f3007
Use ≈
nalimilan Sep 11, 2022
93b3d97
Review fixes
nalimilan Oct 8, 2022
fa2e5e8
Merge remote-tracking branch 'origin/master' into nl/tableregressionm…
nalimilan Oct 8, 2022
5fb340d
More fixes
nalimilan Oct 21, 2022
1925039
Update test/runtests.jl
nalimilan Oct 21, 2022
21a047c
Merge remote-tracking branch 'origin/master' into nl/tableregressionm…
nalimilan Nov 20, 2022
c262432
Merge remote-tracking branch 'origin/nl/tableregressionmodel' into nl…
nalimilan Nov 20, 2022
5c6bf20
Add deprecation to access formula
nalimilan Nov 20, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
Docs, etc.
  • Loading branch information
@nalimilan
nalimilan committed Jun 19, 2022
commit c244147bc8321979d358273ff8516e623d0fd3ee
4 changes: 2 additions & 2 deletions docs/src/api.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@ The most general approach to fitting a model is with the `fit` function, as in
julia> using Random

julia> fit(LinearModel, hcat(ones(10), 1:10), randn(MersenneTwister(12321), 10))
LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, LinearAlgebra.CholeskyPivoted{Float64, Matrix{Float64}}}}:
LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, CholeskyPivoted{Float64, Matrix{Float64}}}}:

Coefficients:
────────────────────────────────────────────────────────────────
Expand All @@ -41,7 +41,7 @@ This model can also be fit as
julia> using Random

julia> lm(hcat(ones(10), 1:10), randn(MersenneTwister(12321), 10))
LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, LinearAlgebra.CholeskyPivoted{Float64, Matrix{Float64}}}}:
LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, CholeskyPivoted{Float64, Matrix{Float64}}}}:

Coefficients:
────────────────────────────────────────────────────────────────
Expand Down
32 changes: 8 additions & 24 deletions docs/src/examples.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,9 +20,7 @@ julia> data = DataFrame(X=[1,2,3], Y=[2,4,7])
3 │ 3 7

julia> ols = lm(@formula(Y ~ X), data)
StatsModels.TableRegressionModel{LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, LinearAlgebra.CholeskyPivoted{Float64, Matrix{Float64}}}}, Matrix{Float64}}

Y ~ 1 + X
LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, CholeskyPivoted{Float64, Matrix{Float64}}}}:

Coefficients:
─────────────────────────────────────────────────────────────────────────
Expand Down Expand Up @@ -56,9 +54,7 @@ julia> data = DataFrame(X=[1,2,2], Y=[1,0,1])
3 │ 2 1

julia> probit = glm(@formula(Y ~ X), data, Binomial(), ProbitLink())
StatsModels.TableRegressionModel{GeneralizedLinearModel{GLM.GlmResp{Vector{Float64}, Binomial{Float64}, ProbitLink}, GLM.DensePredChol{Float64, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}}}, Matrix{Float64}}

Y ~ 1 + X
GeneralizedLinearModel{GLM.GlmResp{Vector{Float64}, Binomial{Float64}, ProbitLink}, GLM.DensePredChol{Float64, Cholesky{Float64, Matrix{Float64}}}}:

Coefficients:
────────────────────────────────────────────────────────────────────────
Expand Down Expand Up @@ -97,9 +93,7 @@ julia> quine = dataset("MASS", "quine")
131 rows omitted

julia> nbrmodel = glm(@formula(Days ~ Eth+Sex+Age+Lrn), quine, NegativeBinomial(2.0), LogLink())
StatsModels.TableRegressionModel{GeneralizedLinearModel{GLM.GlmResp{Vector{Float64}, NegativeBinomial{Float64}, LogLink}, GLM.DensePredChol{Float64, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}}}, Matrix{Float64}}

Days ~ 1 + Eth + Sex + Age + Lrn
GeneralizedLinearModel{GLM.GlmResp{Vector{Float64}, NegativeBinomial{Float64}, LogLink}, GLM.DensePredChol{Float64, Cholesky{Float64, Matrix{Float64}}}}:

Coefficients:
────────────────────────────────────────────────────────────────────────────
Expand All @@ -115,9 +109,7 @@ Lrn: SL 0.296768 0.185934 1.60 0.1105 -0.0676559 0.661191
────────────────────────────────────────────────────────────────────────────

julia> nbrmodel = negbin(@formula(Days ~ Eth+Sex+Age+Lrn), quine, LogLink())
StatsModels.TableRegressionModel{GeneralizedLinearModel{GLM.GlmResp{Vector{Float64}, NegativeBinomial{Float64}, LogLink}, GLM.DensePredChol{Float64, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}}}, Matrix{Float64}}

Days ~ 1 + Eth + Sex + Age + Lrn
GeneralizedLinearModel{GLM.GlmResp{Vector{Float64}, NegativeBinomial{Float64}, LogLink}, GLM.DensePredChol{Float64, Cholesky{Float64, Matrix{Float64}}}}:

Coefficients:
────────────────────────────────────────────────────────────────────────────
Expand Down Expand Up @@ -164,9 +156,7 @@ julia> form = dataset("datasets", "Formaldehyde")
6 │ 0.9 0.782

julia> lm1 = fit(LinearModel, @formula(OptDen ~ Carb), form)
StatsModels.TableRegressionModel{LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, LinearAlgebra.CholeskyPivoted{Float64, Matrix{Float64}}}}, Matrix{Float64}}

OptDen ~ 1 + Carb
LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, CholeskyPivoted{Float64, Matrix{Float64}}}}:

Coefficients:
───────────────────────────────────────────────────────────────────────────
Expand Down Expand Up @@ -213,9 +203,7 @@ julia> LifeCycleSavings = dataset("datasets", "LifeCycleSavings")
35 rows omitted

julia> fm2 = fit(LinearModel, @formula(SR ~ Pop15 + Pop75 + DPI + DDPI), LifeCycleSavings)
StatsModels.TableRegressionModel{LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, LinearAlgebra.CholeskyPivoted{Float64, Matrix{Float64}}}}, Matrix{Float64}}

SR ~ 1 + Pop15 + Pop75 + DPI + DDPI
LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, CholeskyPivoted{Float64, Matrix{Float64}}}}:

Coefficients:
─────────────────────────────────────────────────────────────────────────────────
Expand Down Expand Up @@ -321,9 +309,7 @@ julia> dobson = DataFrame(Counts = [18.,17,15,20,10,21,25,13,13],
9 │ 13.0 3 3

julia> gm1 = fit(GeneralizedLinearModel, @formula(Counts ~ Outcome + Treatment), dobson, Poisson())
StatsModels.TableRegressionModel{GeneralizedLinearModel{GLM.GlmResp{Vector{Float64}, Poisson{Float64}, LogLink}, GLM.DensePredChol{Float64, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}}}, Matrix{Float64}}

Counts ~ 1 + Outcome + Treatment
GeneralizedLinearModel{GLM.GlmResp{Vector{Float64}, Poisson{Float64}, LogLink}, GLM.DensePredChol{Float64, Cholesky{Float64, Matrix{Float64}}}}:

Coefficients:
────────────────────────────────────────────────────────────────────────────
Expand Down Expand Up @@ -378,9 +364,7 @@ julia> round(optimal_bic.minimizer, digits = 5) # Optimal λ
0.40935

julia> glm(@formula(Volume ~ Height + Girth), trees, Normal(), PowerLink(optimal_bic.minimizer)) # Best model
StatsModels.TableRegressionModel{GeneralizedLinearModel{GLM.GlmResp{Vector{Float64}, Normal{Float64}, PowerLink}, GLM.DensePredChol{Float64, LinearAlgebra.Cholesky{Float64, Matrix{Float64}}}}, Matrix{Float64}}

Volume ~ 1 + Height + Girth
GeneralizedLinearModel{GLM.GlmResp{Vector{Float64}, Normal{Float64}, PowerLink}, GLM.DensePredChol{Float64, Cholesky{Float64, Matrix{Float64}}}}:

Coefficients:
────────────────────────────────────────────────────────────────────────────
Expand Down
8 changes: 2 additions & 6 deletions docs/src/index.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -85,9 +85,7 @@ julia> data = DataFrame(y = rand(rng, 100), x = categorical(repeat([1, 2, 3, 4],


julia> lm(@formula(y ~ x), data)
StatsModels.TableRegressionModel{LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, LinearAlgebra.CholeskyPivoted{Float64, Matrix{Float64}}}}, Matrix{Float64}}

y ~ 1 + x
LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, CholeskyPivoted{Float64, Matrix{Float64}}}}:

Coefficients:
───────────────────────────────────────────────────────────────────────────
Expand All @@ -108,9 +106,7 @@ julia> using StableRNGs
julia> data = DataFrame(y = rand(StableRNG(1), 100), x = repeat([1, 2, 3, 4], 25));

julia> lm(@formula(y ~ x), data, contrasts = Dict(:x => DummyCoding()))
StatsModels.TableRegressionModel{LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, LinearAlgebra.CholeskyPivoted{Float64, Matrix{Float64}}}}, Matrix{Float64}}

y ~ 1 + x
LinearModel{GLM.LmResp{Vector{Float64}}, GLM.DensePredChol{Float64, CholeskyPivoted{Float64, Matrix{Float64}}}}:

Coefficients:
───────────────────────────────────────────────────────────────────────────
Expand Down
13 changes: 5 additions & 8 deletions src/glmfit.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -226,10 +226,10 @@ end

abstract type AbstractGLM <: LinPredModel end

mutable struct GeneralizedLinearModel{G<:GlmResp,L<:LinPred,F<:Union{FormulaTerm,Nothing}} <: AbstractGLM
mutable struct GeneralizedLinearModel{G<:GlmResp,L<:LinPred} <: AbstractGLM
rr::G
pp::L
f::F
f::Union{FormulaTerm,Nothing}
nalimilan marked this conversation as resolved.
Show resolved Hide resolved
fit::Bool
end

Expand Down Expand Up @@ -504,7 +504,6 @@ function fit(::Type{M},
data,
d::UnivariateDistribution,
l::Link=canonicallink(d);
# TODO: support passing wts and offset as symbols
offset::Union{AbstractVector, Nothing} = nothing,
wts::Union{AbstractVector, Nothing} = nothing,
dofit::Bool = true,
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

if they're calling fit instead of a constructor, I kinda feel like we shouldn't add the complexity of dofit kwarg.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah that doesn't make a lot of sense, but it's supported currently so I'd say we'd better keep supporting it. I've added deprecations so that we can remove them at some point.

Expand All @@ -517,11 +516,9 @@ function fit(::Type{M},
throw(DimensionMismatch("number of rows in X and y must match"))
end

# TODO: allocate right type upfront
yf = float(y)
off = offset === nothing ? similar(yf, 0) : offset
wts = wts === nothing ? similar(yf, 0) : wts
rr = GlmResp(yf, d, l, off, wts)
off = offset === nothing ? similar(y, 0) : offset
wts = wts === nothing ? similar(y, 0) : wts
rr = GlmResp(y, d, l, off, wts)
res = M(rr, cholpred(X), f, false)
return dofit ? fit!(res; fitargs...) : res
end
Expand Down
9 changes: 6 additions & 3 deletions src/linpred.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -237,7 +237,10 @@ end
function modelframe(f::FormulaTerm, data, contrasts::AbstractDict)
Tables.istable(data) ||
throw(ArgumentError("expected data in a Table, got $(typeof(data))"))
data, _ = StatsModels.missing_omit(Tables.columntable(data), f)
t = Tables.columntable(data)
msg = StatsModels.checknamesexist(f, t)
msg != "" && throw(ArgumentError(msg))
data, _ = StatsModels.missing_omit(t, f)
sch = schema(f, data, contrasts)
f = apply_schema(f, sch, LinPredModel)
f, modelcols(f, data)
Expand Down Expand Up @@ -299,8 +302,8 @@ function StatsBase.predict(mm::LinPredModel, data;
lower = Vector{Union{Float64, Missing}}(missing, nr)
upper = Vector{Union{Float64, Missing}}(missing, nr)
tup = (prediction=view(prediction, nonmissinginds),
lower=view(lower, nonmissinginds),
upper=view(upper, nonmissinginds))
lower=view(lower, nonmissinginds),
upper=view(upper, nonmissinginds))
predict!(tup, mm, new_x; kwargs...)
return (prediction=prediction, lower=lower, upper=upper)
end
Expand Down
8 changes: 5 additions & 3 deletions src/lm.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,17 +75,19 @@ residuals(r::LmResp) = r.y - r.mu
"""
LinearModel

A combination of a [`LmResp`](@ref) and a [`LinPred`](@ref)
A combination of a [`LmResp`](@ref), a [`LinPred`](@ref),
and possibly a `FormulaTerm`

# Members

- `rr`: a `LmResp` object
- `pp`: a `LinPred` object
- `f`: either a `FormulaTerm` object or `nothing`
"""
struct LinearModel{L<:LmResp,T<:LinPred,F<:Union{FormulaTerm,Nothing}} <: LinPredModel
struct LinearModel{L<:LmResp,T<:LinPred} <: LinPredModel
rr::L
pp::T
f::F
f::Union{FormulaTerm,Nothing}
end

LinearAlgebra.cholesky(x::LinearModel) = cholesky(x.pp)
Expand Down
8 changes: 4 additions & 4 deletions test/runtests.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -730,8 +730,8 @@ end
gm11, newX, interval=:confidence, interval_method=:delta) ==
predict(gm11, newX, interval=:confidence, interval_method=:delta)
@test predict!((prediction=similar(Y, size(newX, 1)),
nalimilan marked this conversation as resolved.
Show resolved Hide resolved
lower=similar(Y, size(newX, 1)),
upper=similar(Y, size(newX, 1))),
lower=similar(Y, size(newX, 1)),
upper=similar(Y, size(newX, 1))),
gm11, newX, interval=:confidence, interval_method=:transformation) ==
predict(gm11, newX, interval=:confidence, interval_method=:transformation)
@test_throws ArgumentError predict!((prediction=similar(Y, size(newX, 1)),
Expand Down Expand Up @@ -819,8 +819,8 @@ end
mm, newX, interval=:confidence) ==
predict(mm, newX, interval=:confidence)
@test predict!((prediction=similar(Y, size(newX, 1)),
lower=similar(Y, size(newX, 1)),
upper=similar(Y, size(newX, 1))),
lower=similar(Y, size(newX, 1)),
upper=similar(Y, size(newX, 1))),
mm, newX, interval=:prediction) ==
predict(mm, newX, interval=:prediction)
@test_throws ArgumentError predict!((prediction=similar(Y, size(newX, 1)),
Expand Down